South Africa’s biggest forum. Discuss, discover, and connect with thousands of members.
Imagine typing hey here its already one man hour lost and then you get all the retards to chime in...
Orchestration in what manner? My GPT 5.5 performs perfectly.
I think they mean you have Fable as a master agent, then that spins up subagents with different models to do bits of work.
Bit of overkill to use a high level agent as a orchestrator. Orchestrator is effectively a router, sure use a high agent for review but the actual orchestration can easily be done by lower otherwise you burn a ton of tokens unnecessarily.
Nah, I would think of it more as like a project manager. So have a high level review agent but it can delegate the basic task handling to a lower agent which just assigns and moves the work around. If you have the high level as the central, you will burn tokens doing mundane. If you have one that handles interactions but high level reviews and signs off, less burn. A orchestrator is just that, its job is to coordinate inputs and outputs, not necessarily work on the actual task.
You need the most intelligent model as an Orchestrator. The aim is to pawn off work to cheaper models and it only reviews the output.
SA is a working example of not putting the most intelligent in charge
Yeah, anyone who does that needs a frontal lobotomy.
Saves Fable tokens tbh and Fable has only actually been reviewing my plans and the actual implementations. GPT-5.5, K2.7 and Opus are writing the code. Need to test GLM5.2 still.
Yeah, I use 5.5 high/xhigh for everything. Ain’t got no time to worry about limits. I’m making a billion $$ SaaS here (by that I mean vapourware, that’ll die in my GitHub) .Saves Fable tokens tbh and Fable has only actually been reviewing my plans and the actual implementations. GPT-5.5, K2.7 and Opus are writing the code. Need to test GLM5.2 still.
Only finished my Fable allotment on my $20 Claude account and not going to go to $100
